You are here

Information Criteria Bug?

3 posts / 0 new
Last post
bwiernik's picture
Offline
Joined: 01/30/2014 - 19:39
Information Criteria Bug?

I'm fitting a pseudo-bifactor model to a covariance matrix. When I run summary(), I get the following fit statistics:

observed statistics:  45 
estimated parameters:  22 
degrees of freedom:  23 
fit value ( -2lnL units ):  -3093.63 
saturated fit value ( -2lnL units ):  -3226.707 
number of observations:  1131 
chi-square:  X2 ( df=23 ) = 133.0765,  p = 1.748705e-17
Information Criteria: 
      |  df Penalty  |  Parameters Penalty  |  Sample-Size Adjusted
AIC:       87.07654               177.0765                       NA
BIC:      -28.63319               287.7554                 217.8771
CFI: 0.9552982 
TLI: 0.930032   (also known as NNFI) 
RMSEA:  0.06505073  [95% CI (0.05250108, 0.07801077)]
Prob(RMSEA <= 0.05): 0.009698167

The AIC/BIC values seem wrong. Hand calculations say that AIC should be -3093.63 + 2*23 = -3047.63. Is there a reason it's ending up with the value of 87.07654?

Edit: Per this thread (https://openmx.ssri.psu.edu/thread/89), OpenMx computes AIC, etc. using the X^2 value, rather than -2LL. This isn't what is described in in the help file for mxSummary (it shows the formula as using -2LL). Perhaps the documentation could be updated?

AdminRobK's picture
Offline
Joined: 01/24/2014 - 12:15
agreed

Agreed--the documentation should be clarified.

mhunter's picture
Offline
Joined: 07/31/2009 - 15:26
Oops

Sorry! This is my fault. I wrote the documentation without verifying our actual behavior. I've updated the documentation now to reflect what we actually do.

https://github.com/OpenMx/OpenMx/commit/c155e04dfb1b1ec6ad6cef00a9f86e7e3eb95abb